3.61
#...
·
5
likes
|
|---|
|
Given the time spent scratching my head on this one, I thought sharing my experience might spare others a few headaches. When two years ago I started to drizzle the subs taken with my refiltered EOS6D, I was always getting stars that looked like in Image 1 (top left), which shows a typical example at x4 resolution with 50% droplet size. This image and the following ones are relatively noisy because made of a small number of subs for heavy drizzle (typically 100) under varying and non ideal conditions not to waste clear & dark nights. The fainter stars appear peanut shaped with two lobes along a diagonal (e.g. orange circle), while the brighter ones have a more complex shape with a dark core (e.g. red circle).  Image 1 (top left) was acquired with a Zuiko 180 mm/F2.8 and a 6D refiltered with an Astrodon L filter. My initial hypotheses were: 1. Vibrations 2. Tracking errors 3. Pathologic PSF of the lens 4. Bug in implementation of drizzle algorithm 5. Refiltering 1 & 2 could explain elongated stars (although for 2 the effect should be in RA only), but not different aspects for fainter and brighter stars. 2 was also unlikely because I was getting the same with my AZ-EQ6 and with a staradventurer mini (elongated stars visible here or here at the native resolution). 3 could not explain different shapes for different stars either, and I saw the same peanut stars when I switched to a Samyang 135. 4 was a good candidate since I wrote my own implementation in Python, but I ended up excluding that possibility also based on validations tests on simulated images. I was thus left with 5 - an issue due to the refiltering - but since I had the 6D modded when I bought it, I did not have reference images to compare against. I thus rented a (stock) 6D to test and the first images I got (Image 2) exhibited nicely round stars (e.g. blue circle), but still a dark core in the bright ones (red circle). The stars were also wider than the minor axis of the elongated stars that I was getting with the modded 6D. I then realized that I was facing two separate problems: 1. A "star eater" effect causing the dark cores 2. Duplicate star images possibly from the refiltering I was shooting at 6400 ISO because that setting was giving me the smallest read noise from photon transfer curve measurements. I then rolled back to 3200 ISO, and the dark cores disappeared (Image 3, green circle) in all conditions (Images 4, 5, 6). The 6D therefore applies some sort of despiking at ISO > 3200. While I read extensively about the "star eater" of Sony cameras, I could not find a reference to something similar on the 6D. Interestingly, the images do not show dark cores at the native resolution (inset in Image 1) while the 45° elongation is visible. This may be why I did not find any reference to this with 6Ds. Remained the elongated star images, which I convinced myself were actually duplicates. The smoking gun was that the two lobes were always at 45 degrees and separated by 1 pixel, whatever the optics or mount used (e.g. Images 1 & 4). I also never encountered this issue with an ASI178MC or an ASI178MM). All this pointed to an issue with the sensor. I finally understood what was going on when I learned about Low Pass Filters in DSLRs. LPFs are used to degrade the resolution of the sensor to 2 pixels so that images are never undersampled even if the PSF of the optics used is smaller than that. To first order, each LPF duplicates and shifts the input image by 1 pixel, resulting in 4 superimposed images for two filters oriented at 90 degrees. This prevents moiré effects in photographs of periodic patterns (fences, text, fabric, etc.) but is useless in astro images, especially if drizzle is used to increase the effective resolution of the sensor. The original modification consisted in replacing the second LPF with an Astrodon L filter, while leaving the first LPF (which also serves as a piezo dust cleaner) in place. This results in an anisotropic system which creates anisotropic images. Once convinced that I had the explanation, I had my 6D re-modified by removing the first LPF completely (thus removing the anti-dust piezo function). The Astrodon L filter that replaced the second LPF was also changed to a thicker Astronomik L2, but that should have no influence on the shape of stars. The first test in that configuration (Image 6, to be compared with images 3 & 5) is noisy (made through clouds) but immediately showed round and tight stars (e.g. cyan & green circles), without duplicates. I wish I had known about LPFs at the time of the original modification but everything makes sense now. The stars with the original modification were narrower than those taken with the stock 6D on one axis only because of the removal of one of the two LPFs. The effect is all the more visible as the optics is sharp compared to the pixel pitch. If the PSF of the optics if much smaller than the pixel, the effective PSF of the system is dominated by the PSF of the detector, i.e. the duplicate images. If the PSF of the optics is wider than 2 pixels, the effect can go unnoticed. I'm happy now with a full frame DSLR that supports drizzle. Not all DSLRs are equal and I understand that there are different types of LPFs and correspondingly different types of modifications. Maybe some of these modifications do not create these artifacts. I suppose that I did not search enough and/or correctly because I did not find references to this. I am likely not the only one to drizzle with a 6D having a single LPF removed, which I understand is a pretty common modification of a common DSLR. In any case, I'm interested in getting feedback from others who may have encountered this(these) issue(s). And hopefully this lengthy writeup will be useful to some  CS, Frédéric |
0.00
#...
·
2
likes
|
|---|
|
Frederic, the amount of work you have done benefits all of us. I'm not currently doing astrophotos but I've had a 6D and now have a 6dMk 2. My astrophoto days are behind me, (I'm 73, done with this, let others do it). I keep up with the community, and PROPS for your hard work! Arthur |
